aims - measuring profits and net worth of microenterprises: a field test of eight proxies

8/14/2019 AIMS - Measuring Profits and Net Worth of Microenterprises: A Field Test of Eight Proxies

1/65

Assessing the Impact ofMicroenterprise Services (AIMS)

Management Systems International600 Water Street, S.W.

Washington, D.C. 20024-2488Tel: (202) 484-7170 Fax: (202) 488-0754

E-mail: [email protected]

MEASURING PROFITS AND NET WORTHOF MICROENTERPRISES:

A FIELD TEST OF EIGHT PROXIESNovember 1999

Submitted to:

Monique Cohen, Ph.D.

Office of Microenterprise Development

Global Bureau, USAIDSubmitted by:

Lisa Daniels, Ph.D.Department of EconomicsWashington College

This work was funded by the Microenterprise Impact Project (PCE-0406-C-00-5036-00) of USAIDs Office ofMicroenterprise Development. The Project is conducted through a contract with Management Systems International,in cooperation with the Harvard Institute for International Development, the University of Missouri, and the SmallEnterprise Education and Promotion Network.


2/65

ACKNOWLEDGMENTSI would like to thank several people for their help in carrying out this study. First, Elizabeth Dunn(University of Missouri) developed the original scope of work, made arrangements for the fieldwork,and provided excellent comments to strengthen the two papers (Daniels 1999 and this paper) that

were part of this two-phase project. She also coordinated an advisory group that was asked to makecomments on the first phase of the project. The advisory group included Carolyn Barnes(Management Systems International), Peter Little (University of Kentucky), and Barbara MkNelly(Freedom from Hunger). Like Elizabeth, the members of the advisory group provided excellentcomments and guidance throughout the planning and writing phases of the first paper. MoniqueCohen (United States Agency for International Development) followed the work of the advisorygroup and also provided useful input.

Erin Holleran (Management Systems International) and Walter Ushe (Research InternationalZimbabwe) did an excellent job of coordinating the fieldwork. Their demand for accuracy andconstant supervision led to a high-quality data set. Albert Shuramatongo, one of the supervisors, alsoproved to be invaluable to the team effort because of his reliability and consistency. Other membersof Research International Zimbabwe were also very helpful in coordinating the fieldwork. Tendai

Kadenhe-Mhizha (Managing Director) and Patson Gasura (Quantitative Director) provided excellentmanagement and technical support for the project while Tov Manene (Information TechnologyManager) provided his expertise to guide the data cleaning and verification. Talkmore Zenda alsodid a fine job of data entry.

I am also grateful to the USAID mission in Zimbabwe for allowing us to conduct the study inZimbabwe. In particular, Tichona Mushayandebvu and Martin Hanratty provided important supportto this project. Finally, special thanks are due to the enumerators and supervisors who carried outthe survey. Any remaining errors in the report are the authors responsibility.

Supervisors Albert ShuramatongoWalter Ushe

Enumerators Gilbert ChamunorwaMoreblessing ChirondaJuliet GumbieSamuel KapiridzaQuishe MapandaIbrahim MudalaSibusisiwe MusevenziLovemore NdongweEdson Sithole

ii


3/65

EXECUTIVE SUMMARY

Purpose of Paper

This paper represents the second phase of a two-part project to examine alternative measures of theprofits and net worth of microenterprises. Because full measures of profits and net worth are very

difficult and expensive to collect, donors and practitioners tend to look for other variables, such aschanges in sales revenue or changes in the value of fixed assets, to assess the impact ofmicroenterprise support programs. While these measures offer some indication of the changes inan enterprises status, profits and net worth are much better indicators of enterprise growth andstability. The first phase of this project developed eight proxies to measure profits and net worth anddesigned a field test to examine these proxies. This paper presents the results of the field test.

Survey Method

The field test to evaluate the profit and net worth proxies was conducted in Zimbabwe from Augustto September, 1999. A microenterprise was defined as any type of income-generating activity orbusiness that sold at least 50 percent of its output and employed up to three workers. The definitionof workers included the proprietor, unpaid family members, paid workers, and apprentices. A total

of 448 questionnaires were administered in one urban area and one smaller town. These enumerationareas were selected randomly using a stratified, one-stage cluster sampling technique.

Criteria for Judging Proxies

The proxies were judged by two criteria: accuracy and cost. Accuracy was measured by severalmethods: (1) the percentage of cases that could be estimated by proprietors; (2) the ease with whichproprietors answered the questions related to each proxy; (3) the percentage of cases with positiveprofits; (4) the level of variation within each proxy as compared to the other proxies; and (5) thecorrelation of each proxy with the other measures. Cost was measured by the time needed toimplement each proxy. Obviously, there may be tradeoffs between these two criteria since a greaterlevel of accuracy may require a greater number of questions.

Definitions of the Proxies

Four profit proxies and four net worth proxies were measured along with a full measure for eachvariable. The simplest profit proxy was based on a single question asking the proprietor to estimateprofits for the last month. The second profit proxy was based on three questions that asked for thevalue of the product consumed by the household, money from the enterprise used by the household,and any money left over. The third profit proxy used a more traditional approach of asking for salesin the past month followed by a list of operating expenses and the amount spent on each. The fourthproxy examined sales over the last year as well as operating costs and depreciation costs. A separatesection for traders was also used to examine the costs of restocking the business. Finally, the fullmeasure of profit included all of the components of the fourth proxy as well as information on theoutput consumed or given away, sharing of business assets with the household, and detailed labor

information.

The simplest net worth proxy was based on a single question that asked the proprietor for an estimateof net worth at the time of the interview. The second proxy was based on the value of fixed assetsif they were to be sold today. The information on fixed assets was combined with the value ofinventory, accounts receivable, and outstanding debt to estimate the third proxy. The fourth proxyincluded all of the components of the third proxy plus the cash of the business. This was done by

iii


4/65

asking how much the proprietor could invest if he or she had a good opportunity. Finally, the fullmeasure of net worth included all of the components of the fourth proxy with more detailedinformation about the inventory and sharing of fixed assets with the household. A more directquestion about the amount of cash-on-hand for the business was also included.

Implementation and Statistical Analysis of the Profit Proxies

The time to administer the questions for each of the profit proxies ranged from less than one minuteto 15 minutes. The single-question proxy had the highest proportion of cases that could not beestimated and it was the most difficult for the proprietors. While the first two proxies had no caseswith negative profits, the proportion of cases with negative estimates increased as the proxiesbecame more complex. Although economic theory indicates that firms will operate in the short runwith negative profits, the third, fourth, and full measure of profits produced negative profits in one-third to one-half of all cases. It is unlikely that such a large number of firms operate with negativereturns. Overall, the large proportion of negative estimates indicate that the more complex measuresof profits are not accurate.

The third proxy, based on sales and costs in the last month, showed the greatest degree of variationcompared to the other proxies indicating that it is an inaccurate measure of profits. The Pearson

correlation coefficients revealed that the first two proxies were positively correlated and the twomost complex measures were positively correlated. The correlation between the simple measuresand the complex measures, however, was negative. The rank correlation with Kruskal-Wallis testsshowed the same results.

Implementation and Statistical Analysis of the Net Worth Proxies

The time to administer the questions for each of the net worth proxies ranged from less than oneminute to eight minutes. As in the case of the profit proxies, the single-question proxy had thehighest proportion of cases that could not be estimated and it was the most difficult proxy for theproprietors. There were a few cases with negative net worth values among the full measure and thethird and fourth proxies. This is not necessarily inaccurate, however, since some enterprises mayhave large outstanding debts.

The Pearson correlation coefficients and the rank correlations with the Kruskal-Wallis test showedthat all of the proxies were positively correlated. Overall, the net worth proxies appeared to bepossible substitutes for the full measure of net worth.

Correlation Between Profit and Net Worth

In addition to examining the net worth and profit proxies separately, Pearson correlation coefficientsshowed that there was a positive relationship between the first two profit proxies and all of the networth proxies. The more complex measures of profit, however, were negatively correlated with thenet worth proxies. These results strengthen the conclusion that the two simplest measures of profitsare more accurate than the most complex measures of profits.

Conclusions

The results from this paper indicate that the single-question proxies for profits and net worth are toodifficult for proprietors to answer and result in a large number of cases that cannot be estimated.Among the more complex measures, a greater degree of complexity in the profit proxies leads to lessaccurate results with a large proportion of negative estimates. Furthermore, the more complex

iv


5/65

measures are negatively correlated with the simpler measures. The third proxy, based on sales andoperating costs over the last month, appears to be the least reliable estimate. It has the highestcoefficient of variation and it is positively correlated with the simpler proxies in some analyses andpositively correlated with the more complex measures in other analyses. Overall, the second proxybased on three questions appears to be the best measure of profits. All proprietors could answer thequestions related to this proxy and it did not produce any negative profits. Furthermore, this proxy

is positively correlated with the net worth measures.

In the case of the net worth proxies, all of the measures appear to produce accurate results. Theproxies exhibit relatively similar coefficients of variation and they are positively correlated with eachother. Nonetheless, the third proxy showed the highest correlation with the full measure of networth. In addition to being most closely correlated with the full measure, this proxy is relativelyquick to implement and it avoids the sensitive questions related to the cash of the business.

v


6/65

TABLE OF CONTENTSACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiEXECUTIVE SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiI. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1II. SURVEY METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2III. DEFINITIONS AND IMPLEMENTATION OF PROFIT PROXIES . . . . . . . . . . . . . . . 3

A. Definitions of Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3B. Implementation of Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

IV. RESULTS FOR PROFIT PROXIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8A. Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8B. Cumulative Density Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9C. Pearson Correlation Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10D. Rank Correlations and the Kruskal-Wallis Test . . . . . . . . . . . . . . . . . . . . . . . . . 10E. Implications for the Measurement of Microenterprise Profits . . . . . . . . . . . . . . . 12

V. DEFINITIONS AND IMPLEMENTATION OF NET WORTH PROXIES . . . . . . . . . . 12A. Definitions of Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12B. Implementation of Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

VI. RESULTS FOR NET WORTH PROXIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15A. Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16B. Cumulative Density Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16C. Pearson Correlation Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17D. Rank Correlations and the Kruskal-Wallis Test . . . . . . . . . . . . . . . . . . . . . . . . . 18E. Relative Magnitude of Difference Between

the Full Measure and the Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19F. Implications for the Measurement of Enterprise Net Worth . . . . . . . . . . . . . . . . 19

VII. CORRELATION BETWEEN PROFITS AND NET WORTH . . . . . . . . . . . . . . . . . . . 20VIII. CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

vi


7/65

APPENDICES

APPENDIX 1: Percent of Cases That the Proprietor Did Not Knowthe Answer or Refused to Answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

APPENDIX 2: End-of-Survey Questionnaire Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29APPENDIX 3: Sensitivity Comments Provided by Enumerators . . . . . . . . . . . . . . . . . . . . . . . . 32APPENDIX 4: Rank Correlation Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36APPENDIX 5: Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42APPENDIX 6: A Comparison of Sales to Profit Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

TABLES IN TEXT

Table 1: Implementation of Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6Table 2: Descriptive Statistics for Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8Table 3: Pearson Correlation Coefficients for Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . 10Table 4: Kruskal-Wallis Test Results for Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11Table 5: Implementation of Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15Table 6: Descriptive Statistics for Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16Table 7: Pearson Correlation Coefficients for Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . 18Table 8: Kruskal-Wallis Test Results for Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . 18Table 9: Relative Magnitude of Differences Between Net Worth Proxies and Full Measure . . . 19Table 10: Pearson Correlation Coefficients between the Profit and Net Worth Proxies . . . . . . . 20

TABLES IN APPENDIX 6

Table A.1: Sales Last Month Compared to the Profit Measures . . . . . . . . . . . . . . . . . . . . . . . . . 58Table A.2: Pearson Correlation Coefficients for Profit Proxies and Sales . . . . . . . . . . . . . . . . . 58Table A.3: Kruskal-Wallis Test Results for Sales and Profit Proxies . . . . . . . . . . . . . . . . . . . . . 59

FIGURES IN TEXT

Figure 1: Cumulative Density Functions of the Profit Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . 9Figure 2: Cumulative Density Functions of the Net Worth Proxies . . . . . . . . . . . . . . . . . . . . . . 17

FIGURES IN APPENDIX 4

Figure A.1: Rank Correlation, Profit Proxy 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Figure A.2: Rank Correlation, Profit Proxy 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37Figure A.3: Rank Correlation, Profit Proxy 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Figure A.4: Rank Correlation, Profit Proxy 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Figure A.5: Rank Correlation, Full Measure of Profits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Figure A.6: Rank Correlation, Net Worth Proxy 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Figure A.7: Rank Correlation, Net Worth Proxy 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Figure A.8: Rank Correlation, Net Worth Proxy 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40Figure A.9: Rank Correlation, Net Worth Proxy 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Figure A.10: Rank Correlation, Full Measure of Net Worth . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

vii


8/65


9/65

I. INTRODUCTION

Information on enterprise profits and net worth can be critical to assessing the impact ofmicroenterprise services. Because full measures of profits and net worth can be difficult andexpensive to collect, there is a tendency to look for alternative variables, such as changes in salesrevenue or changes in the value of fixed assets, to assess the impact of microenterprise support

programs. While these alternative measures offer some indication of the changes in an enterprisesstatus, profits and net worth are considered to be much better indicators of enterprise growth andstability.

This paper represents the second phase of a two-part project to examine alternative measures ofmicroenterprise profits and net worth.1 The first phase of the project designed a field test anddeveloped eight proxies to measure profits and net worth based on a review of previous studies(Daniels 1999). In particular, that report identified over twelve methods used by different studiesto calculate profits. These methods ranged from an estimate of profits last month provided by theproprietor to the more complex methods of subtracting capital services and the value of non-familylabor from value added. Examining the individual components of profits, there were nine methodsto estimate sales, seven methods to estimate labor costs, seven methods to estimate operating costs,and six methods to estimate fixed costs. Overall, a total of 378 proxies could be developed based

on the various combinations of the components to estimate profits. In the case of net worth, nostudies were located that estimated the complete value of net worth. This was partly due to thesensitivity of questions related to net worth, such as questions about the cash-on-hand of theenterprise.

The second phase of this project included a field test of the eight proxies identified in the first phaseand an analysis of the results, which are presented in this report. The definitions of the proxies usedin this study are provided in sections III and V below. They range from single-question proxies tofull measures for each variable including up to 209 and 59 subquestions for profits and net worth,respectively. The proxies were judged by two criteria: accuracy and cost. Accuracy is measured byseveral methods: (1) the percentage of cases that could be estimated by proprietors; (2) the ease withwhich proprietors answered the questions related to each proxy; (3) the percentage of cases withpositive profits; (4) the level of variation within each proxy as compared to the other proxies; and(5) the correlation of each proxy with the other measures. Cost was measured by the time neededto implement each proxy. Obviously, there may be tradeoffs between these two criteria since agreater level of accuracy may require a greater number of questions.

Overall, the results show that the simplest proxies appear to provide more accurate estimates ofprofits, whereas the more complex methods produce a large proportion of negative estimates.Although some firms do operate with negative profits in the short run, the high proportion of caseswith negative estimates indicates that these measure are not very accurate. Among the net worthmeasures, all of the proxies appear to be possible substitutes for the full measure of net worth.Nonetheless, the third proxy based on fixed assets, inventory, accounts receivable, and outstandingdebt appeared to be the best proxy. All proprietors could answer the questions related to this proxyand it had the highest correlation with the full measure of net worth.

1 Both phases of this project were conducted as part of the Assessing the Impact of Microenterprise Services(AIMS) Project. The goals of the AIMS project are to gain a better understanding of the processes by whichmicroenterprise services strengthen businesses and improve the welfare of microentrepreneurs and their households.In addition, the goal of the AIMS project is to improve the ability of USAID and its partners to assess the impacts oftheir microenterprise programs. More information on the AIMS Project is available on the website(http://www.mip.org).

1


10/65

A key limitation of this study is that the full measure of profits is based on a single-visit survey.Ideally, the profit proxies should be compared to a full measure that is estimated through repeatedvisits. Furthermore, the full measure of profits turned out to be inaccurate because of the largeproportion of negative estimates, as mentioned above. The profit proxies are not, therefore, judgedby their correlation to the full measure of profits. In the case of net worth, a single visit isappropriate to estimate a full measure since the concept of net worth is associated with a single point

in time.

This paper begins, in section II, with a brief description of the survey methods used in the study.Section III provides some basic information about the profit proxies including the questions used,the time needed to collect the data for each proxy, the number of cases that could not be estimated,the number of cases with negative profits, and the level of difficulty and sensitivity for each proxy.The statistical analysis of the profit proxies is presented in section IV. The types of statistics includecoefficients of variation, Pearson correlation coefficients, and rank correlations. Sections V and VIrepeat the same information for the net worth proxies. The correlation between the profit proxiesand net worth proxies is then examined in section VII. Finally, section VIII offers some briefconclusions.

II. SURVEY METHOD

For the purposes of this survey, a microenterprise was defined as any income-generating activity withthree or fewer workers selling 50 percent or more of its product. Agriculture, mining, and forestrywere excluded from the survey. A total of 448 questionnaires were administered in Zimbabwe fromAugust 23 to September 2, 1999. The sample selection and data collection methods are describedbriefly below.

In 1991, 1993, and 1998, the USAID-funded GEMINI project conducted national surveys ofmicroenterprises in Zimbabwe (McPherson 1991; Daniels 1994; McPherson 1998). Each time, thesurveys used a stratified, one-stage cluster sampling technique. This involved three steps. First, thecountry was divided into eight strata based on population density and commercial activities. Urbanareas were defined as cities with more than 20,000 inhabitants as estimated by the 1982 census.Within this group, there were four strata: high-density areas, low-density areas, commercial districts,and industrial areas.2 The remaining four strata in rural areas included small towns, growth points,district councils, and rural councils.3 Second, a random sample of enumeration areas within eachstratum was selected. The enumeration areas were based on areas delineated by the Central StatisticsOffice for the national census. Third, all households in each selected enumeration area wereapproached. If a household had an enterprise, a questionnaire was administered. In addition, allmobile businesses and businesses located outside of households were included in the surveys.

A subset of enumeration areas from the GEMINI surveys was selected randomly to be included inthe survey for this study. In particular, 230 proprietors were identified and interviewed in oneenumeration area from the urban high-density stratum and 218 proprietors were identified andinterviewed in one enumeration area from the smaller town stratum. Based on these sample sizes,

2 High-density areas are typically inhabited by low-income households while low-density areas are inhabitedby high-income households.

3 Growth points are towns designated by the government to promote rural development. Incentives areprovided in these towns to promote the establishment and growth of businesses. For more information on growth pointssee Pedersen (1992), Gasper (1988), and Wekwete (1987).

2


11/65

the results from the urban high-density area represent a 17 percent margin of error with a 90 percentconfidence level for the mean value of profit. In the smaller town area, the results represent a 21percent margin of error with a 90 percent confidence level.

The data were collected by nine enumerators and two supervisors. Enumerators and supervisorswere trained for one week, followed by the final field tests of the questionnaire. Twelve enumerators

attended training, but only nine were selected for the fieldwork based on written test scores andperformance during training.

To administer the survey, enumerators visited all houses, shops, street vendors, and hawkers withinthe geographic boundaries of each enumeration area. Questionnaires were then coded and checkedfor errors. Each enumerator checked his or her own work at the end of the day and was then askedto check the work of one other enumerator. The supervisors then checked all questionnaires andgave them to the data entry person who also checked them for errors. As an extra measure ofaccuracy, the data entry person entered the data from each questionnaire twice. Once all data wereentered, frequency charts were examined for each variable and any unusual numbers were identifiedand returned to the enumerator for verification.

III. DEFINITIONS AND IMPLEMENTATION OF PROFIT PROXIES

A. Definitions of Profit Proxies

As described in the introduction, information was collected to estimate four profit proxies and fournet worth proxies. The definitions and a brief description of each profit proxy are provided below.Because wages for the proprietor and any unpaid employees were not deducted from the four profitsproxies, all of the proxies represent the return to proprietors and unpaid workers. Only the fullmeasure of profits deducts the value of in-kind payments to unpaid employees. The full measure,therefore, represents returns to the proprietor only.

Profit Proxy 1: Profits in last month as estimated by the proprietor in a single question

The first proxy was based on a single question. Proprietors were asked to estimate their profits overthe past week or month. They were reminded to consider all costs such as transport, inputs, supplies,and paid labor. If the proprietor gave the profits for the last week only, the enumerator asked ifprofits were low, average, or high for that week. An estimate for the month was then recorded bymultiplying the response by four if the week was average or adding the profit for each week if itvaried over the past month.

Profit Proxy 2: Value of product consumed plus money from the enterprise used by thehousehold plus any money left over

The second proxy was based on three questions used by the World Bank as part of the LivingStandards Measurement Surveys (LSMS).4 The first question asked proprietors to estimate the value

of the product normally consumed by the household. The second question asked proprietors toestimate how much money from the business they normally use for themselves or their household.Finally, the third question asked proprietors to estimate the amount of money that they had left over

4 The World Bank has conducted Living Standard Measurement Surveys in several dozen countries. Thestudies are used to examine household income and expenditure patterns. For a review of the questions related tomicroenterprises from these studies see Vijverberg and Mead (forthcoming).

3


12/65

after consuming some of the product and using some of the money from the business. Convertingthese answers to monthly estimates and adding them together provided the second proxy for profits.One advantage of this measure is that it avoids estimation of sales, fixed assets, and operating costsplus all of the recall problems associated with these components of profits.

Profit Proxy 3: Sales revenue minus operating costs in the last month

The third proxy was based on five questions with a maximum of 28 subquestions. Profits wereestimated as sales revenue minus operating costs in the last month. Information on sales revenuewas collected in a single question that asked about sales last month. Operating costs were based ona list of costs and the amount spent on each per week or month in the last month. Although thisproxy approaches a full measure of profits, it does not include depreciation of fixed assets. It alsoignores seasonality of sales throughout the year.

Profit Proxy 4: Sales revenue minus operating costs minus depreciation in the last year

The fourth proxy for profits was estimated as sales revenue minus operating costs and depreciation.It was based on a total of seven questions with a maximum of 138 subquestions. Information onsales revenue was based on the average amount earned in high, low, and medium months. 5

Information on operating costs was collected through the same list of expenses described above forthe third profit proxy. In addition, however, a ratio of variable costs to sales revenue was estimatedfor the past month and then applied to high, low, and average sales months to determine the costsper month throughout the year. Depreciation of fixed assets was also incorporated into this proxyby subtracting 20 percent of the current value of equipment and five percent of the current value ofbuildings.6 Finally, a separate section was used for traders to estimate the costs to restock theirbusinesses. After estimating the annual profits using this proxy, the number was converted to amonthly estimate in order to compare it to the other proxies.

Full Measure of Profits: Proxy 4 plus output consumed by the household or given away andrefinements in depreciation, labor use, and asset sharing

The full measure of profits was based on nine questions with a maximum of 209 subquestions. In

addition to all of the information used in the fourth profit proxy, the full measure includedinformation about output consumed or given away by the household and detailed information onindividual workers employed by the microenterprise over the past year. Rather than using a straight-line depreciation method as in the fourth proxy, the full measure first estimated the proportion ofeach asset used by the business and then depreciated that portion based on the number of years leftof use as estimated by the proprietor.

In theory, the full measure should provide the most accurate estimate of profits. As described in theintroduction, however, the full measure had a large proportion of cases with negative estimates.Again, while some firms may have negative profits, it is unlikely that such a large proportion offirms operate with negative profits. The full measure was not, therefore, used as a standard to

5 Proprietors were asked whether each month of the year was a high, low, or medium sales month. The numberof each type of month was then multiplied by the average sales in that type of month as stated by the proprietor in orderto determine annual sales. There was no assumption that all proprietors had a certain number of high, low, or mediumsales months.

6 The value of buildings was only included if it was purchased for the business. If the business was run fromthe home, the cost of the house was not included.

4


13/65

examine the other proxies. Other characteristics, listed in the introduction, were used to judge eachproxy.

5


14/65

B. Implementation of Profit Proxies

Table 1: Implementation of Profit Proxies

Profit Profit Profit Profit FullProxy 1 Proxy 2 Proxy 3 Proxy 4 Measure

Question numbers from the C1 C6, C7, C8 B5, C3, B5, D2, B5, D2, E1,questionnaire (see appendix 6) D1, E1, F1 E1, F1, F1, G1, G2,

G1, G2, G3, H1, J1H1

Number of questions includingmaximum subquestions

1 3 28 138 209

Average time to collect proxy perinterview (minutes)

0.9 1.8 7.5 13.7 15.2

% of cases that could not beestimated due to missing information

32% 0% 14% 11% 17%

% of cases with negative estimates(among those that answered)

0% 0% 37% 55% 52%

Average level of difficulty(0=none, 3=extreme)

1.3 0.73 0.44 0.50 0.56

Table 1 illustrates the questions used for each of the four profit proxies and the full measure of profit.As illustrated, the time to administer each proxy ranged from less than one minute to 15 minutes forthe full measure. The simplest proxies were completed in under two minutes whereas the mostcomplex proxies required eight or more minutes per interview on average.

Although the first proxy was the simplest in terms of the number of questions, it had the highest

proportion of cases that could not be estimated. Close to one-third of proprietors could not answerthe single question for this proxy. Alternatively, all proprietors answered the questions related tothe second proxy. Among the two most complex proxies and the full measure of profits, 11 to 17percent could not be estimated. The table in appendix 1 details the extent of proprietors inabilityor refusal to answer the individual questions involved in each proxy.

As described in the AIMS report for the first phase of this project (Daniels 1999), negative profitestimates are common among the more complex measures of profits. For example, Vijverberg andMead (forthcoming) showed that the percent of cases with negative profits in the LSMS data rangedfrom 14 percent in Vietnam to 64 percent in Ghana. They suggest that the large percentage ofnegative cases is not plausible. In this field test in Zimbabwe, the two simplest measures of profitsdid not yield any cases with negative profits. As the proxies became more complex, however, thepercentage of negative cases ranged from 37 percent to 55 percent. While some firms may operate

with negative profits in the short run, one-third to one half of all firms operating with negative profitsseems unrealistic. A closer examination of the negative estimates revealed that the greatest numberof negative estimates are generated from cases where input costs are greater than sales. 7 Similarly,

7 For the third proxy, input costs, operating costs, and restocking costs (for traders) were greater than sales in43 percent, 33 percent, and 13 percent of the negative cases, respectively. For the fourth proxy, input costs, operatingcosts, traders costs, and depreciation costs were greater than sales in 36 percent, 18 percent, five percent, and two

6


15/65

all firms operating with positive profits, as indicated by the first two proxies, seems unrealistic.

The level of difficulty that proprietors experienced when answering questions was estimated throughan end-of-survey questionnaire administered to the enumerators. The enumerators were asked to rateeach question according to the following scale: 0=no difficulty; 1=some difficulty; 2=a lot ofdifficulty; and 3=an extreme amount of difficulty. The level of difficulty does not refer to sensitivity

of the question. Instead, the level of difficulty refers to the ability of the proprietor to provide theinformation. Considering the modal values, the highest level of difficulty for any single questionrelated to the profit proxies was a one or some difficulty. As the number of questions per proxyincreased, the number of questions with a mode of one increased as well. This indicates that theproxies themselves are not necessarily more difficult, but there are a greater number of more difficultquestions as the proxies become more complex. The table in appendix 2 provides more detail aboutthe level of difficulty for each of the questions used in the proxies.

Sensitivity issues were covered by written comments provided by the enumerators. The completeset of comments are included in appendix 3. In general, all of the enumerators reported that thequestions concerning cash or profits were sensitive. Considering only those questions related to theprofit proxies, seven of the ten enumerators mentioned the estimate of profits last month (questionC1) as one of the most sensitive questions. The questions related to wages paid to employees was

also mentioned as sensitive by seven of the enumerators. One enumerator reported that it wasparticularly difficult for the proprietor to answer these questions with more than one employeepresent during the interview.

percent of the negative cases, respectively. Finally, for the full measure of profits, inputs costs, operating costs(excluding labor), labor costs, fixed costs, restocking costs, and depreciation were greater than sales in 37 percent, 12percent, five percent, four percent, four percent, and 0.5 percent of the negative cases, respectively.

7


16/65

IV. RESULTS FOR PROFIT PROXIES8

This section reports the results of the statistical analysis related to the proxies and full measure ofprofits. Prior to the analysis, extreme outliers were removed based on the assumption that theyprovided inaccurate data. In particular, all cases that were three standard deviations above or belowthe mean for any proxy were removed from the data set. A total of 20 cases, or 4.5 percent, were

removed.9

A. Descriptive Statistics

Table 2: Descriptive Statistics for Profit Proxies

Mean (Z$) Median (Z$) StandardDeviation (Z$)

Coefficient ofVariation (%)

Profit Proxy 1 1885 900 2754 146%

Profit Proxy 2 2615 1500 3241 124%

Profit Proxy 3 1448 285 17758 1226%

Profit Proxy 4 -1096 -35 5343 488%

Full Measure -948 -19 4949 522%The means were tested for pairwise differences using the Wilcoxon test. All pairs showed statistically significantdifferences. The medians were tested for pairwise differences using a chi-square statistic. All pairs showedstatistically significant differences with the exception of the fourth proxy and the full measure.

Table 2 lists the mean, median, standard deviation, and coefficient of variation for the proxies andfull measure of profit. The coefficient of variation, which provides a measure of the variability ofeach proxy in percentage terms, is measured as follows:

Standard DeviationCoefficient of Variation = x 100

Mean

The first two proxies have remarkably similar characteristics and the lowest coefficients of variationamong the five measures. The third proxy, based on sales last month, has the greatest coefficient ofvariation. Although it is impossible to determine which level of variation among the proxies is themost accurate, the large level of variation within the third proxy as compared to the other proxies

8 The analyses reported in this section as well as information on the implementation of the profit proxies were

also examined at the sector level (manufacturing, commerce, and service) and at the stratum level (urban and rural).Because there were no substantially different results than those reported at the aggregated level, the tables were notincluded in this paper. If the reader is interested in these tables, they are available from the author.

9 Since the point of this study was to determine which proxies provide the most accurate estimates of profits,it could be argued that it is not appropriate to remove any cases. If, however, the extreme outliers remain in the dataset, the Pearson correlation coefficients and the descriptive statistics would be almost useless. In addition, studies thatattempt to measure profits with more complex methods will most likely produce extreme outliers that would be removedfrom the data set before analysis.

8


17/65

10 Figure 1 has been truncated in order to provide a clearer illustration of the density functions.

9

suggests that this proxy may not be as accurate. The fourth proxy and the full measure had similarcharacteristics. This is not surprising since the calculation of the two measures is very similar .

B. Cumulative Density Functions

Figure 1 shows the cumulative density functions of the five profit proxies. 10 Ideally, these

distributions should be identical since the proxies attempt to estimate the same number. Asillustrated on the graph, however, the distribution of the first two proxies is quite different than theremaining proxies due to the large number of negative estimates for the third, fourth, and fullmeasures of profits. Considering only the first two proxies, the distributions are quite similar. Thedistribution of the second profit proxy, however, suggests a higher estimate of profits than the firstprofit proxy. The fourth and full measures of profits appear to be almost identical. As mentionedabove, this is not surprising since the calculation of the two measures is similar.

0

10

20

30

40

50

60

70

80

90

100

-10000 -8000 -6000 -4000 -2000 0 2000 4000 6000 8000 10000

Profits

Cumulative%o

fallcases

Profit Proxy 1Profit Proxy 2

Profit Proxy 3

Profit Proxy 4

Full Measure of Profit

Figure 1: Cumulative Density Functions of the Profit Proxies

C. Pearson Correlation Coefficients

The Pearson correlation coefficients for the proxies and the full measure of profits are provided in

table 3. In cases where the coefficients are statistically significant, they can be interpreted as thestrength or weakness of the linear association between two variables. The extreme values ofnegative one or positive one indicate a perfect negative or positive correlation between two variables,respectively.


18/65

Examining the individual pairs of relationships in table 3, the first proxy is positively correlated withthe second and third proxies. The first proxy is not, however, correlated with the fourth and fullmeasures of profits. The second proxy is significantly correlated with third, fourth, and full measureof profits. The magnitude of the correlation is, however, very weak and it is negative in the case ofthe fourth and full measures of profits. The fourth and full measures are strongly correlated, whichis not surprising since the calculation of the two measures is very similar. Overall, the results show

that the first two proxies could possibly be substituted for one another and the first and third proxy.The fourth and full measures of profits do not have a strong relationship with any of the three simplermeasures. Because of the large number of negative cases among the most complex measures, itappears that the simplest measures may be more accurate. There is a tradeoff, however, since manyproprietors had difficulty answering the questions related to the single-question proxy.

Table 3: Pearson Correlation Coefficients for Profit Proxies

Proxy 2 Proxy 3 Proxy 4 FullMeasure

Proxy 1 .615* .615* -.094 -.061

Proxy 2 .476* -.141* -.144*

Proxy 3 .196* .176*

Proxy 4 .961*

*Significant at the .10 level.

D. Rank Correlations and the Kruskal-Wallis Test

In addition to examining Pearson correlation coefficients, the correlations between the proxies andfull measure were tested using rank correlations with the Kruskal-Wallis test. This test comparesthe mean rank of one variable within the deciles of another variable. 11 By replacing the profitestimates with their ranks, this test eliminates the influence of extreme outliers. The results of theKruskal-Wallis test are reported in table 4, which shows a significant relationship between allcombinations of the proxies. The only exception is that the first two proxies are not significantlyrelated to the fourth and full measures of profits.

11 For the Kruskal-Wallis test, two proxies are compared by creating a variable that is a decile of the firstproxy. The lowest ten percent of the values of the first proxy are given a value of one. The second lowest ten percentof the values are given a value of two, and so on. This first variable ranges from one to ten. A second variable is thencreated that replaces the value of the second proxy by its rank in the data set from one to 428 (the number of cases inthe data set). The mean rank of the second variable is then examined within the deciles created for the first proxy. Thenull hypothesis is that the mean rank for the second variable is the same in all ten deciles of the first variable. If thereis a relationship between the two variables, the null hypothesis is rejected.

10


19/65

Table 4: Kruskal-Wallis Test Results for Profit Proxies

Decile Grouping Enterprise Profits

Proxy 1 Proxy 2 Proxy 3 Proxy 4 Full Measure

Proxy 1 Chi-SquareAsymp. Sig.

70.03.000

21.12.012

5.64.776

6.12.728


70.82.000

18.60.029

10.93.280

10.03.348


52.02.000

57.50.000

74.96.000

62.75.000


57.42.000

57.49.000

60.82.000

302.50.000

Full Measure Chi-SquareAsymp. Sig.

62.35.000

55.98.000

45.11.000

303.32.000

The Chi-square is significant at the 90 percent confidence level if the asymptotic significance is less than .10.

The results of the Kruskal-Wallis test only indicate if there is a significant relationship between two

proxies. The test does not indicate if the relationship is positive or negative. To examine this issue,figures A.1 through A.5 in appendix 5 illustrate the shape of each relationship. The horizontal axisshows the decile for one proxy while the vertical axis shows the mean rank of the remaining proxies.An upward slope in the graphs means that there is a positive correlation, whereas a downward slopeindicates a negative correlation. A flat slope indicates little or no correlation.

Figure A.1 shows the relationship between the first-proxy deciles and the mean rank for the othermeasures of profits. The relationship between the first and second proxy appears to be positive. Therelationship between the first proxy and the remaining measures, however, is much less clear. FigureA.2 uses the deciles of the second proxy on the horizontal axis. Again, only the first two proxieshave a positive relationship. When examining the mean rank of the proxies within the deciles of thethird, fourth, and full measure of profits, figures A.3, A.4, and A.5 show almost identical patterns.The first two proxies exhibit a U-shaped line whereas the last three measures appear to be positivelycorrelated.12 Finally, there is a positive correlation between the first two proxies and the fourth andfull measures in the higher deciles or the positive estimates of profits for the fourth and full measure.Overall, these results are similar to the Pearson correlation coefficient results. There appears to bea positive relationship between the first two proxies and a positive relationship among the third,fourth, and full measure of profits. The first two proxies, however, do not show a positivecorrelation with the three other measures of profit.

E. Implications for the Measurement of Microenterprise Profits

Combining the information on the implementation of the proxies and the statistical analyses, theresults indicate that the first and second profit proxies appear to be better estimates of profits thanthe more complex measures. In terms of cost, the simpler proxies are quicker to implement. In

terms of accuracy, the first two proxies did not exhibit the large number of negative cases foundamong the more complex proxies. Again, however, it is unrealistic to assume that there are no firms

12 The U-shaped pattern of the first two proxies could indicate that among those cases with large negativevalues for the fourth and full measures, proprietors estimated their own profits in the first two proxies at much higherrates. At the fifth decile, when profits are estimated as zero for the fourth and full measures, the estimates of proxiesone and two are much lower.

11


20/65


21/65

Net Worth Proxy 2: Current value of fixed assets

The second proxy of net worth was based on the value of fixed assets. A list of 20 fixed assets wasread to the proprietor, who was asked to estimate the value of each item if it were to be sold that day.

Net Worth Proxy 3: Current value of fixed assets plus inventory, accounts receivable, and

outstanding debt

In addition to the value of fixed assets, the third proxy included the value of the current inventory,accounts receivable, and outstanding debt for a total of 32 subquestions. The value of currentinventory was estimated as the total value of raw materials and the total value of finished products.Accounts receivable were estimated as the total amount owed today by customers, traders, andfamily members or friends. Similarly, outstanding debt was estimated by reading a list of possiblesources of debt to the proprietor and asking for the amount still owed to each source as of the dayof the interview.

Net Worth Proxy 4: Proxy 3 plus cash of business (opportunity to invest)

The fourth proxy included all of the components of the third proxy. In addition, it included the cash

available to the business today. This was done by asking how much the proprietor could spend fromthe business cash and savings if she or he had an excellent opportunity for a business investment onthe day of the interview.

Full Measure of Net Worth: Current value of fixed assets (portion used by business) plus detailedinventory value, accounts receivable, outstanding debt, and cash-on-hand of business

The full measure was based on a total of 59 subquestions and included all of the components of thefourth proxy with slightly greater detail. The value of inventory, for example, was calculated byasking for the quantity of every item in stock and the value of the item if it were to be sold today.The value of fixed assets was calculated using the same list described for the second proxy.Proprietors were also asked, however, if the asset was shared by another business or the householdand the proportion of the time that the asset was actually used by the business. Only the proportion

of the asset used by the business was incorporated into the value of net worth. Finally, proprietorswere asked for the amount of cash-on-hand today instead of asking about the cash available for aninvestment opportunity.

B. Implementation of Net Worth Proxies

Table 5 lists the questions used for each of the four proxies and for the full measure of net worth.As illustrated in table 5, questions for the simplest net worth proxies were completed in under threeminutes whereas the most complex proxies required an average of seven to eight minutes perinterview. Although the first proxy was the simplest in terms of the number of questions, over one-third of all proprietors could not estimate their net worth in this way. In contrast, all proprietorsanswered the questions related to the second and third proxies. For the fourth proxy and the full

measure of net worth, four percent and 13 percent could not be estimated, respectively. The tablein appendix 1 provides more detail about the percentage of cases that could not be estimated forindividual questions included in the proxies.

The percentage of cases with negative estimates of net worth was quite low for the two mostcomplex proxies and the full measure of net worth. It is reasonable to expect some businesses tohave a negative net worth since they may have considerable debt. All of the cases had positiveestimates for the first two net worth proxies.

13


22/65

As described earlier, the level of difficulty that proprietors experienced when answering eachquestion was estimated by the enumerators following the survey. A level of zero indicated nodifficulty whereas a level of three indicated an extreme amount of difficulty. Considering the modalresponse, the first proxy, based on one question, had a mode of three, indicating that this wasextremely difficult to answer. The list of fixed assets for the second proxy had a modal response ofzero, indicating that this is a relatively easy set of questions for the proprietor. For the third, only

one question had a modal value of one and all others were zero. Similarly, the fourth proxy, whichadds one question to the third proxy, had zero as a modal value for all questions with the exceptionof two questions with a modal value of one. The full measure had primarily zeroes for the modalvalues with the exception of two questions with a mode of one and one question (cash of thebusiness) with a mode of three. Overall, these results indicate that the first proxy is the least accuratein terms of the ability of the proprietor to answer the question. The full measure had one questionthat was extremely difficult for the proprietors. The majority of the questions for the full measureand the second, third, and fourth, proxies, however, could be answered without much difficulty. Thetable in appendix 2 provides more detail regarding the level of difficulty for the individual questionsincluded in the proxies.

Table 5: Implementation of Net Worth Proxies

Proxy 1 Proxy 2 Proxy 3 Proxy 4 FullMeasure

Question numbers from the C5 H1 H1, I1, I2, H1, I1, H1, I3,questionnaire K1, K2, I2, K1, K1, K2,

K3, K4 K2, K3, K3, K4,K4, L1 L2, L3

Number of questions includingsubquestions

1 20 32 33 59

Average time to collect proxyper interview (minutes)

0.9 2.6 7.0 7.6 7.6

% of cases that could not be 36% 0% 0% 4% 13%estimated due to missinginformation

% of cases with negative 0% 0% 4.5% 2.3% 2.8%estimates (among those thatanswered)

The enumerators provided a set of written comments about sensitivity issues, which are reported inappendix 3. In general, there were many more questions related to the net worth proxies that wereconsidered sensitive compared to the profit proxies. In particular, enumerators identified thequestions related to the detailed inventory, outstanding debts, and savings (questions I3, K1, K3, K3,

L1, L2, and L3) as sensitive questions. Overall, the question related to cash-on-hand (L2) appearedto be the most sensitive.

14


23/65

VI. RESULTS FOR NET WORTH PROXIES13

This section reports the results related to the proxies and full measure of net worth. As describedearlier, twenty cases were removed from the data set because they included measures of profit or networth that were more than three standard deviations away from the mean of the proxy. In addition,the detailed inventory used for the full measure of net worth had many estimates that were

abnormally high. An examination of the questionnaires revealed that this was due to incorrectrecording of sales units. For those cases where the detailed inventory was 100 times greater than theestimate of the value of inventory provided by the proprietor, the detailed inventory value wasreplaced by the proprietors estimate.14

A. Descriptive Statistics

Table 6 provides the means, medians, standard deviations, and coefficients of variation for theproxies and for the full measure of net worth. Unlike the profit proxies, which exhibited coefficientsof variation ranging from 124 percent to 1,226 percent, the range for the coefficients of variation forthe net worth proxies is much smaller. Also, as described earlier, the second and third proxiesattempt to measure only a portion of the full measure of net worth. Their means should, therefore,be lower than then mean of the full measure. The fourth measure, which uses the full value of assets

used by the business rather than a the portion actually used, should exhibit a higher mean value thanthe full measure.

Table 6: Descriptive Statistics for Net Worth Proxies

Mean (Z$) Median (Z$) StandardDeviation (Z$)

Coefficient ofVariation (%)

Net Worth Proxy 1 12450 3000 27141 218%

Net Worth Proxy 2 3565 130 13558 380%

Net Worth Proxy 3 10181 1680 24955 245%

Net Worth Proxy 4 21167 5090 54840 259%

Full Measure 12147 2978 25883 213%The means were tested for pairwise differences using the Wilcoxon test. All pairs showed statistically significantdifferences with the exception of the first and third proxy. The medians were tested for pairwise differences using achi-square statistic. All pairs showed statistically significant differences.

B. Cumulative Density Functions

Figure 2 shows the cumulative density functions of the five net worth proxies. As described above,the second and third measures of net worth should provide lower estimates of net worth.

13 The analyses reported in this section as well as information on the implementation of the net worth proxieswere also examined at the sector level (manufacturing, commerce, and service) and at the stratum level (urban and rural).Because there were no substantially different results than those reported at the aggregated level, the tables were notincluded in this paper. If the reader is interested in these tables, they are available from the author.

14 This type of error could be avoided in future surveys by providing enumerators with more thorough trainingon the recording of inventory units.

15


24/65

15 Figure 2 has been truncated in order to provide a clearer illustration of the density functions.

16

Alternatively, the first net worth proxy should exhibit the same distribution as the full measure sincethese two measures both estimate the full value of net worth. Finally, the fourth proxy should exhibitslightly higher estimates of net worth than the full measure. These two measures are almost identicalwith the exception of the calculation of fixed assets. The fourth proxy uses the full value of fixedassets whereas the full measure uses only the portion of the asset actually used by the business. Allof these patterns are exhibited in Figure 2.15 The distributions of the full measure and the first net

worth proxy are very similar. The second, third, and fourth measures offer successively higherestimates of net worth as expected and the fourth proxy offers a higher estimate than the full measureof net worth.

0

10

20

30

40

50

60

70

80

90

100

-1000 1000 3000 5000 7000 9000 11000 13000 15000

Net Worth

Cumulative%o

fallcases

Net Worth Proxy 1

Net Worth Proxy 2

Net Worth Proxy 3

Net Worth Proxy 4

Full Measure of Net Worth

Figure 2: Cumulative Density Functions for Net Worth Proxies

C. Pearson Correlation Coefficients

The Pearson correlation coefficients for the proxies and the full measure of net worth are providedin table 7. Because the five measures are not attempting to estimate the exact same value, thecorrelations become more important as a means of judging accuracy. All pairs of proxies and thefull measure are positively correlated and these correlations are statistically significant. Thissuggests that all of the proxies work reasonably well with the exception of the second and fourthproxy where the correlation is relatively weak. The highest degree of correlation is between the thirdproxy and the full measure.


25/65

Table 7: Pearson Correlation Coefficients for Net Worth Proxies

Proxy 2 Proxy 3 Proxy 4 Full Measure

Proxy 1 .569* .705* .509* .706*

Proxy 2 .646* .348* .553*

Proxy 3 .578* .838*

Proxy 4 .561*


D. Rank Correlations and the Kruskal-Wallis Test

Table 8: Kruskal-Wallis Test Results for Net Worth Proxies

DecileGrouping

Proxy 1 Proxy 2 Proxy 3 Proxy 4 FullMeasure

Proxy 1 Chi-Square 99.08 160.39 158.10 122.78Asymp. Sig. .000 .000 .000 .000




Full Chi-Square 129.27 382.27 244.52 212.23

Measure Asymp. Sig. .000 .000 .000 .000

The correlation between the measures was tested using rank correlations and the Kruskal-Wallis test,as described earlier. Again, outliers have less influence in this test since the data values are replacedby their ranks. Table 8 shows that there is a significant relationship between each pair of net worthmeasures. This relationship is positive in all cases as illustrated in figures A.6 through A.10 inappendix 5. Overall, this indicates that all of the proxies are appropriate substitutes for the fullmeasure of net worth.

E. Relative Magnitude of Difference Between the Full Measure and the Proxies16

This section examines the relative magnitude of variation within the net worth proxies. In

particular, table 9 shows the percent of cases for each proxy that are two or three times greater or less

16 This analysis was not carried out for the profit proxies since the full measure of profits proved to beunreliable.

17


26/65

than the full measure of net worth.17 Overall, this table shows that there are very few cases thatdeviate substantially from the full measure by a large amount. Only the fourth proxy showed thatclose to one-fifth of the cases produced a much higher estimate of net worth than the full measure.This reflects the fact that the fourth proxy includes the entire value of fixed assets as part of the networth, whereas the full measure uses a reduced portion of the value of the asset if it is shared withother businesses or with the household.

F. Implications for the Measurement of Microenterprise Net Worth

Combining the information on the implementation of the proxies and the statistical analyses, theresults indicate that the third proxy appears to be the best estimate of net worth for a number ofreasons. First, it is less sensitive than the most complex measures because it avoids asking about thecash of the business. In terms of difficulty, enumerators indicated that only two questions posedsome difficulty for the proprietors, whereas the majority of the questions posed no difficulty at all.In terms of the statistical analyses, the third proxy had the highest correlation with the full measureof net worth. Finally, all proprietors could answer the questions related to this proxy. It should bekept in mind, however, that the third proxy is only a partial measure of net worth. It omits the valueof the cash-on-hand of the business. Although it is positively correlated with net worth, it willunderstate the true value of net worth.

Table 9: Relative Magnitude of Differences Between Net Worth Proxies and Full Measure

Ratio of Proxy Net Worth Net Worth Net Worth Net Worthto Full Measure Proxy 1 Proxy 2 Proxy 3 Proxy 4

3 times greater 7.5% 0.3% 3.2% 19.0%3 times less 0.8% 0.0% 0.5% 1.1%

2 times greater 11.3% 0.3% 5.1% 30.7%2 times less 0.8% 0.0% 0.8% 1.1%

VII. CORRELATION BETWEEN PROFITS AND NET WORTH

Although net worth is measured at one point in time (i.e., net worth at the time of the interview) andprofits are measured over some previous time period (e.g., last month or last year) there could besome correlation between the two measures. For example, a firm that earns high profits may reinvestthat profit into the business and thus exhibit higher net worth. Obviously this correlation will dependon the extent to which proprietors reinvest profits into the business. Because this relationship mayexist, this section examines the correlation between the two sets of proxies. Table 10 shows theresults. The first and second profit proxies are positively correlated with all of the net worthmeasures. The correlation, however is very weak in some cases. The third profit proxy exhibitsmore irregular results. The correlation is only statistically significant for the first, third, and fullmeasure of net worth and the correlation in these cases is very weak. The two most complex

measures of profit are negatively correlated with the net worth measures. Because all of the networth proxies produced more consistently accurate estimates, these results strengthen the conclusionthat the two simplest measures of profits are more accurate than the most complex measures of

17 This assumes that the full measure of net worth is the most accurate measure. Although there is no way toprove that the full measure is the most accurate method without extensive data collection, the results from the analysisabove indicate that all of the proxies and the full measure of net worth are relatively reliable.

18


27/65

profit.

Table 10: Pearson Correlation Coefficients Between the Profit and Net Worth Proxies

ProfitProxies

Net Worth Proxies

Proxy 1 Proxy 2 Proxy 3 Proxy 4 Full Measure

Proxy 1 .551* .210* .528* .374* .566*

Proxy 2 .321* .204* .356* .203* .354*

Proxy 3 .226* .001 .121* .059 .152*

Proxy 4 -.156* -.084* -.206* -.174* -.171*

Full Measure -.165* -.126 -.237* -.185* -.213*


VIII. CONCLUSIONS

The results related to the profit proxies can be summarized as follows:

The simplest profit proxy had the highest number of cases that could not be estimated by theproprietor.

The second profit proxy could be estimated by all proprietors and it did not produce the largenumber of negative estimates as in the case of the complex proxies. Furthermore, it waspositively correlated with the net worth proxies. Nonetheless, it was somewhat sensitive forproprietors.

The third profit proxy, based on sales and costs last month, appeared to provide the mostinconsistent estimate of profits. In some analyses it was correlated with the simpler proxies,and in other analyses it was correlated with the more complex measures. It also produceda large number of negative cases and it had an extremely high coefficient of variationcompared to the other proxies.

The fourth proxy and the full measure of profit produced large numbers of negativeestimates. These profit estimates were negatively correlated with the simpler proxies.

Based on these results, the second profit proxy appears to be the most accurate measure of profitsand it has a relatively low cost of implementation compared to the more complex proxies.

The results related to the net worth proxies can be summarized as follows:

The simplest net worth proxy had the highest number of cases that could not be estimatedand it was extremely difficult for the proprietor to answer.

All proxies appeared to produce accurate results, and they were positively correlated.Although all of the net worth proxies could be used as a substitute for the full measure of net worth,the third proxy showed the highest correlation with the full measure of net worth. In addition, this

19


28/65

proxy is relatively quick to implement and it avoids the sensitive questions associated with the cashof the business that are included in the fourth proxy and the full measure of net worth. Finally, allproprietors could answer the questions related to this proxy.

20


29/65

REFERENCESDaniels, Lisa. 1994. Changes in the Small-Scale Enterprise Sector from 1991 to 1993. Results of

a Second Nationwide Survey in Zimbabwe. GEMINI Technical Report No. 71. Bethesda,MD: Development Alternatives Inc.

Daniels, Lisa. 1999. Alternatives for Measuring Profits and Net Worth of Microenterprises.Technical Note. AIMS Project Report, USAID/G/EG/MD. Washington, D.C.: ManagementSystems International.

Gasper, D. 1988. Rural Growth Points and Rural Industries in Zimbabwe. Ideologies andPolicies.Development and Change. Vol 19, No. 3: 425-266.

McPherson, Michael. 1991. Micro- and Small-Scale Enterprises in Zimbabwe: Results of aCountry-wide Survey. GEMINI Technical Report No. 25. Bethesda, MD: DevelopmentAlternatives, Inc.

McPherson, Michael. 1998. Changes in Zimbabwe's Micro and Small Enterprise Sector,1991-1998: Evidence from a Third Country-wide Survey. GEMINI Technical Report.Bethesda, MD: Development Alternatives, Inc.

Pedersen, Paul Ove. 1992. Agricultural Marketing and Processing in Small Towns in Zimbabwe Gutu and Gokwe. Chapter in The Rural-urban Interface: Expansion and Adaptation, J.Baker and P.O. Pedersen (eds.). Copenhagen, Denmark: The Scandinavian Institute ofAfrican Studies.

Vijverberg, Wim, and Donald Mead. Forthcoming. The Household Enterprise Module. ChapterinDesigning Household Survey Questionnaires for Developing Countries: Lessons from TenYears of LSMS Experience, Margaret Grosh and Paul Glewwe (eds.). Washington, D.C.: TheWorld Bank.

Wekwete, K. 1987. Growth Centre Policy in Zimbabwe: A Focus on District Centers. RUPOccasional Paper No. 7. Harare, Zimbabwe: Department of Rural and Urban Planning,University of Zimbabwe.

21


30/65

APPENDIX 1:

PERCENT OF CASES THATTHE PROPRIETOR DID NOT KNOW

THE ANSWER ORREFUSED TO ANSWER

22


31/65

Variables that are not listed in this table did not have any cases where the proprietor could notanswer or refused to answer.

Question Variable Label Percent of cases where the:

Proprietor

Could NotAnswer

Proprietor

Refused toAnswer

B3A Month Started 22 0

B3B Year Started 0.4 0

B5A Months Operate in Last Year 0.4 0

B5B2 Days Per Month: Average 0.4 0

B5C2 Hours Per Day: Average 0.2 0

C1 Profit: One Question 29.7 1.8

C2 Profit Last Year for MSE> One Year OldEnterprise

47.3 0.4

C3 Sales Last Week/Month 12.3 0.2

C4 Expenses Last Week/Month 9.2 0

C5 Net Worth: One Question 35 1.3

C6 Value of Product/Services used by HH 4.7 0

C7 Value of Money Used by HH 8.7 0

C8 Money Left 9.7 .7

C8A Time Period 39.5 0

D1B Restock in AVERAGE Month 0.7 0

D2A3 Number of Units Sold Last Day/Week/Month 4.7 0

D2A5 Purchase Price Of Product 0.7 0

D2A7 Units of Sales Per Unit of Purchase 2.5 0

D2B3 Number of Units Sold Last Day/Week/Month 3.6 0

D2B7 Units of Sales Per Unit of Purchase 1.8 0

D2C3 Number of Units Sold Last Day/Week/Month 2.7 0

D2C7 Units of Sales Per Unit of Purchase 0.7 0

D2D3 Number of Units Sold Last Day/Week/Month 1.3 0

D2D7 Units of Sales Per Unit of Purchase 0.4 0

23


32/65


ProprietorCould NotAnswer

ProprietorRefused toAnswer

D2E3 Number of Units Sold Last Day/Week/Month 0.7 0D2E7 Units of Sales Per Unit of Purchase 0.7 0

D2F3 Number of Units Sold Last Day/Week/Month 7.6 0

D2F7 Units of Sales Per Unit of Purchase 0.2 0

D2G3 Number of Units Sold Last Day/Week/Month 5.1 0

D2G7 Units of Sales Per Unit of Purchase 5.1 0

D2H3 Number of Units Sold Last Day/Week/Month 4.2 0

D2H7 Units of Sales Per Unit of Purchase 4.2 0

D2I3 Number of Units Sold Last Day/Week/Month 3.1 0

D2I7 Units of Sales Per Unit of Purchase 3.1 0

D2J3 Number of Units Sold Last Day/Week/Month 2.2 0

D2J7 Units of Sales Per Unit of Purchase 2.2 0

F1F1 Cost: Water 0.4 0

F1H1 Cost: Transport of Inputs 0.2 0

F1I1 Cost: Transport of Final Product 0.9 0

F1M1 Cost: Repairs/Service of Machines 0.2 0

F1N1 Cost: Other 0.2 0

G2A Typical High Sales Per Month 2.2 .9

G2B Typical Average Sales Per Month 9.8 1.1

G2C Typical Low Sales Per Month 1.3 0.4

G3A1 Value: Consumption of Output in Household 3.1 0.2

G3D1 Value: Give Away 2.2 0.2

H1A3 Tools: Time Left of Use 2.9 0.2H1A5 Tools: Price if Sold Today 5.1 2.0

H1B3 Tools: Time Left of Use 2.2 0.2

H1B4 Tools: Original Purchase Price 2.7 0

24


33/65




H1B5 Tools: Price if Sold Today 1.3 1.3H1C3 Tools: Time Left of Use 0.2 0.2

H1C4 Tools: Original Purchase Price 1.1 0

H1C5 Tools: Price if Sold Today 1.8 0.2

H1D4 Tools: Original Purchase Price 0.7 0

H1D5 Tools: Price if Sold Today 0.4 0

H1E3 Furnishings: Time Left of Use 0.4 0.2

H1E4 Furnishings: Original Purchase Price 1.8 0

H1E5 Furnishings: Price if Sold Today 1.1 0.2

H1F3 Furnishings: Time Left of Use 0.2 0

H1F4 Furnishings: Original Purchase Price 0.4 0

H1F5 Furnishings: Price if Sold Today 0.4 0

H1G3 Vehicles: Time Left of Use 0.2 0

H1G4 Vehicles: Original Purchase Price 0.2 0

H1G5 Vehicles: Price if Sold Today 0.7 0

H1H3 Machinery/Equipment: Time Left of Use 4.2 0

H1H4 Machinery/Equipment: Original Purchase Price 5.6 0

H1H5 Machinery/Equipment: Price if Sold Today 6.5 0.9

H1I3 Machinery/Equipment: Time Left of Use 1.3 0

H1I4 Machinery/Equipment: Original Purchase Price 1.3 0

H1I5 Machinery/Equipment: Price if Sold Today 1.1 0.2

H1J3 Buildings: Time Left of Use 0.7 0

H1J4 Buildings: Original Purchase Price 1.6 0H1J5 Buildings: Price if Sold Today 1.3 0.2

H1K5 Buildings: Price if Sold Today 0.2 0

H1M4 Other: Original Purchase Price 0.9 0

25


34/65




H1M5 Other: Price if Sold Today 2.9 0H1N4 Other: Original Purchase Price 0.4 0

H1N5 Other: Price if Sold Today 0.2 0

H1O4 Other: Original Purchase Price 0.2 0

H1P1 Other: Time Owned 0.2 0

H1P4 Other: Original Purchase Price 0.2 0

H1Q1 Other: Time Owned 0.2 0

H1R1 Other: Time Owned 0.2 0

I1 Total Value of Raw Materials if Sold Today 6.7 0

I2 Total Value of Finished Products if Sold Today 10.7 0.2

I3A2 Number of Raw Materials in Inventory 0.7 0

I3B2 Number of Raw Materials in Inventory 0.4 0

I3C2 Number of Raw Materials in Inventory 0.2 0

I3D2 Number of Raw Materials in Inventory 0.2 0

I3E2 Number of Raw Materials in Inventory 0.2 0

I3E3 Cost of One Product/Raw Material 0.2 0

J1A2 Number of Months Worked: Past 12 Months 0.2 0

J1A3 Number of Days Per Month 0.2 0

J1A4 Number of Hours Per Day 0.2 0

J1A5 Salary: Amount 0.7 0.2

J1A7 In-Kind Payment: Amount 2.0 0

J1B2 Number of Months Worked: Past 12 Months 0.2 0

J1B3 Number of Days Per Month 0.2 0J1B5 Salary: Amount 0 0.2

J1B7 In-Kind Payment: Amount 0.2 0

J1C2 Number of Months Worked: Past 12 Months 0.2 0

26


35/65




J1C3 Number of Days Per Month 0.2 0J1C5 Salary: Amount 0 0.4

J1D5 Salary: Amount 0.2 0

K1 Amount: Owed by Customers 0.2 0.4

K3 Amount: Owed by Friends/Family Members 0.7 0

L1 Opportunity to Invest: Amount Available 3.8 0

L2 Cash on Hand Today: Amount 2.9 7.4

L3A Bank Savings: Amount 0.4 3.3

L3B Post Office Savings: Amount 0.2 1.6

L3C Savings Club: Amount .2 0

27


36/65

APPENDIX 2:

END-OF-SURVEY QUESTIONNAIRERESULTS

28


37/65

This table is based on a questionnaire administered to the enumerators at the end of the survey. Eachenumerator estimated the number of minutes to administer the questions listed below and the levelof difficulty on a scale of zero to three.

Question Variable Label AverageNumber of

Minutes toAdminister

Percent of enumeratorsthat reported each level

of difficulty (0=none,3=Extreme)

0 1 2 3

B5A Months Operated in Last Year 0.34 100 0 0 0

B5B Days Operated per Month 0.353 80 20 0 0

B5C Hours Operated per Day 0.41 70 30 0 0

C1 Profit Last Month 1.03 0 70 30 0

C2 Profit Last Year 1.343 0 10 60 30C3 Sales Last Week 0.915 10 80 10 0

C4 Expenses Last Week 1.01 40 50 10 0

C5 Net Worth 2.35 0 0 30 70

C6 Household Consumption 0.739 40 50 10 0

C7 Money Used from Business 0.652 50 50 0 0

C8 Money Left From Business 0.66 30 40 30 0

D1 Traders - Amount to Restock Business 0.985 70 30 0 0

D2 Traders - Prices and Volume 3.8 70 30 0 0

E1 Non-traders - Input Costs 2.9 50 50 0 0

F1 Other Operating Expenses 2.75 40 40 20 0

G1 Sales Volume by Month 1.193 40 40 20 0

G2 Sales Revenue 1.09 30 50 20 0

G3 Produce Consumed or Given Away 0.84 60 40 0 0

H1 Fixed Assets 3.0 50 30 20 0

I1 Value of Raw Material 1.675 40 40 20 0

I2 Value of Finished Products 1.55 50 20 20 10

I3 Inventory of Raw Materials 2.65 70 30 0 0

J1 Employment in the Business 0.915 90 0 10 0

29


38/65

Question Variable Label AverageNumber ofMinutes toAdminister

Percent of enumeratorsthat reported each levelof difficulty (0=none,3=Extreme)

0 1 2 3

K1 Amount Customers Owe You 0.643 80 20 0 0

K2 Amount Traders Owe You 0.318 90 0 10 0

K3 Amount Family or Friends Owe You 0.312 90 0 10 0

K4 Credit Still Owed 0.591 90 10 0 0

L1 Amount You Could Invest Today 0.748 10 40 30 20

L2 Cash From Business Today 0.8330 20 30 20 30

L3 Savings From the Business 0.502 40 40 20 0

30


39/65

APPENDIX 3:

SENSITIVITY COMMENTSPROVIDED BY ENUMERATORS

31


40/65

ENUMERATOR ONE

1) Respondents did not find it easy and safe to answer questions that had something to do with cash,especially C1, C2, L2, L3.

2) The question of licenses was also sensitive because they ended up thinking we had something todo with people having licenses when a business activity is carried out.

3) The question of laborers was also sensitive because they thought maybe we were from a tradeunion since some of them cannot afford to pay their laborers the required wages.

ENUMERATOR TWO

1) In conducting this survey I discovered that proprietors did not want to disclose their financialstatus, especially on questions C1, C2, C3, L2, and L3.

2) Some also found question I3 (inventory) sensitive, especially those who did not have a lot ofthings to sell.

3) Most of the small business proprietors do not pay for licenses and they thought we would takethem to the Income Tax Offices.

4)Those who have employees thought we would report to the Trade Union that they were

underpaying their workers.

ENUMERATOR THREE

1) Respondents were reluctant to answer such questions as C1, L2, and L3 which asked for theamounts of money they make (C1 - profit from last month) and cash they had in hand andat the bank (L2 and L3 respectively).

2) Another sensitive question concerns the salaries of both both the proprietors and their workers(for those that had employees). They were very reluctant to disclose their salary amounts(J1).

ENUMERATOR FOUR

1) Those that were interviewed were not comfortable disclosing their profits.2) Question C2 was also sensitive. People were also not comfortable with C8 as interviewees could

not easily disclose how much money they had after household consumption.3) K4 also caused some sensitivity as interviewees could not easily disclose how much they owed

a certain institution.4) L2 was also sensitive. Those interviewed thought that it was part of their secrecy to disclose

moneys that they had in their coffers.5) J11 was also sensitive because proprietors were not comfortable to disclose the salaries of their

employees.ENUMERATOR FIVE

1) Questions that involved money, like C1,C2, C3, and C8, were quite sensitive.2) Also questions concerning savings, like L1, L2 and L3, were sensitive.3) Generally, money, profit, and savings oriented questions were quite sensitive.

ENUMERATOR SIX

32


41/65

1) People did not want to talk about licenses and workers. They thought we were concerned abouttaxes.

2) They also did not want to talk about cash on hand (L2); it was too personal.

ENUMERATOR SEVEN

1) J1 was a sensitive question because they thought we wanted to liaise with labor so they theycould be penalized for paying them too little.

2) L2 was also a sensitive question because they did not want us or strangers to know how muchthey have as cash in hand.

3) L3 was another sensitive question because they did not want us to know how much they have inthe bank or elsewhere.

ENUMERATOR EIGHT

1) The most sensitive questions were those which required the respondents to give us their cashinflows and outflows - especially their savings. C1 and C2, which were asking for the

profitability of the businesses, I believe forced the respondents to make an assumption thatwe wanted to know about their income, which they thought was none of our business.

2) K1 was also sensitive because the respondents thought we wanted to know about their financialposition.

3) Lastly, I think L2 and L3 were the most sensitive questions because the respondents did not trustus and could not believe our purpose.

ENUMERATOR NINE

1) Respondents has difficulties in understanding Net Worth (Question C5).2) On questions C2, C3 and C4, the respondents could not easily recall their usiness operations

during the previous weeks or months.

3) Questions concerning their money from the business were very sensitive. They could not disclosethat. The questions in this category were C1, C2, L2 and L3.

4) On remaining questions, the respondents were able to understand and answered them moreeasily.

ENUMERATOR TEN

1) All questions concerning money were sensitive.2) Question J1 where a proprietor is asked how much money he pays employee one when employee

two is present.3) Question K3, when when a proprietor is asked if friends or family

aims - measuring profits and net worth of microenterprises: a field test of eight proxies

Documents